As from my previous questions, I am trying to build a simple eye tracker. Decided to start from a Linux version (run Ubuntu).

To complete this task one should organize screencasting and webcam capturing in such way that frames from both streams exactly match each other and there is the same number of frames in each of them totally.

Screencasting fsp fully depends on the camera's fsp, so each time we get the image from the webcam we can potentially grab a screen frame and stay happy. However, all the tools for the fast screencasting, like ffmpeg, for example, return the .avi file as the result and require the fsp already known to be started.

From the other side, tools like Java+Robot or ImageMagick seem to require around 20ms to return the .jpg screenshot, which is pretty slow for the task. But they may be requested right after each time the webcam frame is grabbed and provide the needed synchronization.

So the sub-questions are: 1. Does the USD camera's frame rate vary during a single session? 2. Are there any tools which provide fast screencasting frame by frame? 3. Is there any way to make ffmpeg push a new frame to the .avi file only when program initiates this request?

For my task I may either use C++ or Java.

I am, actually, an interface designer, not the driver programmer, and this task seems to be pretty low-level. I would be grateful for any suggestion and tip!